FT-Mamba: A Novel Deep Learning Model for Efficient Tabular Regression

Deep learning has advanced in image, audio, and text, but challenges such as heterogeneity and weak correlations remain in tabular data. This paper presents the FT-Mamba architecture, which merges the capabilities of FT-Transformer with Mamba’s efficiency for effective, scalable sequential data processing. It also uses random augmentation, data balancing, and self-distillation to boost performance. The work provides novel insights and tools for deep learning with tabular data, improving its practical application in real-world scenarios.